DeepSeek V4.1 Flash vs GPT-6 Luna
Same documents, same prompt: where each model gets it right, what it really costs, and how fast it responds.
Verdict
DeepSeek V4.1 Flash is more accurate; GPT-6 Luna is cheaper in 27 of 30 combinations.
More accurate
DeepSeek V4.1 Flash
wins 16 of 30
Cheaper
GPT-6 Luna
cheaper in 27 of 30
Faster
GPT-6 Luna
3.0 sec vs 5.0 sec
Fewer fixes
DeepSeek V4.1 Flash
13 vs 33 documents per 1,000
Across 30 task and document-language combinations (speed and fixes: median).
At a glance
| Accuracy (correct fields) | Per 1,000 documents | |||||
|---|---|---|---|---|---|---|
| Task · documents | DeepSeek V4.1 Flash | GPT-6 Luna | Difference | DeepSeek V4.1 Flash | GPT-6 Luna | Difference |
| Invoice extraction | 99.39% | 99.70% | ▲ 0.31 | $0.52 | $0.51 | −2 % |
| Email triage | 97.63% | 97.48% | ▼ 0.15 | $0.23 | $0.13 | −42 % |
| Personal data detection | 99.62% | 99.62% | ±0 | $0.29 | $0.11 | −63 % |
| Contract clauses | 100.00% | 99.79% | ▼ 0.21 | $0.32 | $0.20 | −37 % |
| Document summaries | 100.00% | 100.00% | ±0 | $0.54 | $0.16 | −71 % |
| Questions about a document | 99.22% | 99.11% | ▼ 0.11 | $0.31 | $0.15 | −51 % |
| Accuracy (correct fields) | Per 1,000 documents | |||||
|---|---|---|---|---|---|---|
| Task · documents | DeepSeek V4.1 Flash | GPT-6 Luna | Difference | DeepSeek V4.1 Flash | GPT-6 Luna | Difference |
| Invoice extraction | 99.90% | 99.28% | ▼ 0.62 | $0.50 | $0.51 | +2 % |
| Email triage | 97.41% | 97.93% | ▲ 0.52 | $0.30 | $0.18 | −40 % |
| Personal data detection | 99.71% | 99.62% | ▼ 0.09 | $0.31 | $0.12 | −62 % |
| Contract clauses | 100.00% | 100.00% | ±0 | $0.42 | $0.27 | −36 % |
| Document summaries | 100.00% | 100.00% | ±0 | $0.78 | $0.21 | −73 % |
| Questions about a document | 100.00% | 100.00% | ±0 | $0.38 | $0.23 | −40 % |
| Accuracy (correct fields) | Per 1,000 documents | |||||
|---|---|---|---|---|---|---|
| Task · documents | DeepSeek V4.1 Flash | GPT-6 Luna | Difference | DeepSeek V4.1 Flash | GPT-6 Luna | Difference |
| Invoice extraction | 99.54% | 99.23% | ▼ 0.31 | $0.57 | $0.54 | −4 % |
| Email triage | 96.37% | 96.96% | ▲ 0.59 | $0.30 | $0.15 | −50 % |
| Personal data detection | 99.81% | 99.62% | ▼ 0.19 | $0.31 | $0.12 | −62 % |
| Contract clauses | 97.08% | 100.00% | ▲ 2.92 | $0.52 | $0.27 | −48 % |
| Document summaries | 100.00% | 100.00% | ±0 | $0.62 | $0.19 | −69 % |
| Questions about a document | 100.00% | 98.00% | ▼ 2 | $0.33 | $0.20 | −39 % |
| Accuracy (correct fields) | Per 1,000 documents | |||||
|---|---|---|---|---|---|---|
| Task · documents | DeepSeek V4.1 Flash | GPT-6 Luna | Difference | DeepSeek V4.1 Flash | GPT-6 Luna | Difference |
| Invoice extraction | 98.21% | 99.13% | ▲ 0.92 | $0.65 | $0.55 | −15 % |
| Email triage | 96.52% | 96.81% | ▲ 0.29 | $0.34 | $0.16 | −54 % |
| Personal data detection | 99.90% | 99.71% | ▼ 0.19 | $0.44 | $0.12 | −73 % |
| Contract clauses | 100.00% | 100.00% | ±0 | $0.40 | $0.26 | −34 % |
| Document summaries | 100.00% | 97.00% | ▼ 3 | $0.88 | $0.19 | −78 % |
| Questions about a document | 100.00% | 99.67% | ▼ 0.33 | $0.35 | $0.19 | −44 % |
| Accuracy (correct fields) | Per 1,000 documents | |||||
|---|---|---|---|---|---|---|
| Task · documents | DeepSeek V4.1 Flash | GPT-6 Luna | Difference | DeepSeek V4.1 Flash | GPT-6 Luna | Difference |
| Invoice extraction | 99.54% | 99.08% | ▼ 0.46 | $0.60 | $0.56 | −6 % |
| Email triage | 96.74% | 97.26% | ▲ 0.52 | $0.34 | $0.16 | −52 % |
| Personal data detection | 99.71% | 99.62% | ▼ 0.09 | $0.37 | $0.12 | −67 % |
| Contract clauses | 99.85% | 99.33% | ▼ 0.52 | $0.45 | $0.27 | −39 % |
| Document summaries | 100.00% | 98.78% | ▼ 1.22 | $0.78 | $0.20 | −75 % |
| Questions about a document | 100.00% | 99.56% | ▼ 0.44 | $0.36 | $0.20 | −43 % |
Difference: GPT-6 Luna vs. DeepSeek V4.1 Flash; accuracy in percentage points, price in percent. The better value is in bold. Click a task to see the details.
Invoice extraction · English (US) (150 documents)
GPT-6 Luna is more accurate by 0.3 pp.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.39% | 99.70% |
| Error-free documents | 93.33% | 96.67% |
| Per 1,000 documents | $0.52 | $0.51 |
| To fix / 1,000 | 67 | 33 |
| Speed (median) | 5.7 sec | 3.4 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Invoice extraction · Czech (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.6 pp.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.90% | 99.28% |
| Error-free documents | 98.67% | 93.33% |
| Per 1,000 documents | $0.50 | $0.51 |
| To fix / 1,000 | 13 | 67 |
| Speed (median) | 5.0 sec | 3.2 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Invoice extraction · German (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.3 pp.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.54% | 99.23% |
| Error-free documents | 94.00% | 91.33% |
| Per 1,000 documents | $0.57 | $0.54 |
| To fix / 1,000 | 60 | 87 |
| Speed (median) | 6.4 sec | 3.7 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Invoice extraction · Spanish (150 documents)
GPT-6 Luna is more accurate by 0.9 pp. GPT-6 Luna is 1.2× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 98.21% | 99.13% |
| Error-free documents | 77.33% | 88.67% |
| Per 1,000 documents | $0.65 | $0.55 |
| To fix / 1,000 | 227 | 113 |
| Speed (median) | 6.3 sec | 3.9 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Invoice extraction · French (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.5 pp. GPT-6 Luna is 1.1× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.54% | 99.08% |
| Error-free documents | 94.00% | 88.00% |
| Per 1,000 documents | $0.60 | $0.56 |
| To fix / 1,000 | 60 | 120 |
| Speed (median) | 6.2 sec | 3.9 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Email triage · English (US) (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.1 pp. GPT-6 Luna is 1.7× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 97.63% | 97.48% |
| Error-free documents | 78.67% | 77.33% |
| Per 1,000 documents | $0.23 | $0.13 |
| To fix / 1,000 | 213 | 227 |
| Speed (median) | 3.8 sec | 2.5 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Email triage · Czech (150 documents)
GPT-6 Luna is more accurate by 0.5 pp. GPT-6 Luna is 1.7× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 97.41% | 97.93% |
| Error-free documents | 77.33% | 81.33% |
| Per 1,000 documents | $0.30 | $0.18 |
| To fix / 1,000 | 227 | 187 |
| Speed (median) | 4.2 sec | 2.6 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Email triage · German (150 documents)
GPT-6 Luna is more accurate by 0.6 pp. GPT-6 Luna is 2× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 96.37% | 96.96% |
| Error-free documents | 68.00% | 72.67% |
| Per 1,000 documents | $0.30 | $0.15 |
| To fix / 1,000 | 320 | 273 |
| Speed (median) | 4.2 sec | 2.8 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Email triage · Spanish (150 documents)
GPT-6 Luna is more accurate by 0.3 pp. GPT-6 Luna is 2.2× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 96.52% | 96.81% |
| Error-free documents | 72.00% | 72.00% |
| Per 1,000 documents | $0.34 | $0.16 |
| To fix / 1,000 | 280 | 280 |
| Speed (median) | 5.5 sec | 3.1 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Email triage · French (150 documents)
GPT-6 Luna is more accurate by 0.5 pp. GPT-6 Luna is 2.1× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 96.74% | 97.26% |
| Error-free documents | 72.67% | 76.00% |
| Per 1,000 documents | $0.34 | $0.16 |
| To fix / 1,000 | 273 | 240 |
| Speed (median) | 5.2 sec | 2.8 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Personal data detection · English (US) (150 documents)
Both are equally accurate. GPT-6 Luna is 2.7× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.62% | 99.62% |
| Error-free documents | 97.33% | 97.33% |
| Per 1,000 documents | $0.29 | $0.11 |
| To fix / 1,000 | 27 | 27 |
| Speed (median) | 3.8 sec | 1.8 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Personal data detection · Czech (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.1 pp. GPT-6 Luna is 2.6× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.71% | 99.62% |
| Error-free documents | 98.00% | 97.33% |
| Per 1,000 documents | $0.31 | $0.12 |
| To fix / 1,000 | 20 | 27 |
| Speed (median) | 4.2 sec | 1.8 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Personal data detection · German (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.2 pp. GPT-6 Luna is 2.6× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.81% | 99.62% |
| Error-free documents | 98.67% | 97.33% |
| Per 1,000 documents | $0.31 | $0.12 |
| To fix / 1,000 | 13 | 27 |
| Speed (median) | 3.7 sec | 2.2 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Personal data detection · Spanish (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.2 pp. GPT-6 Luna is 3.7× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.90% | 99.71% |
| Error-free documents | 99.33% | 98.00% |
| Per 1,000 documents | $0.44 | $0.12 |
| To fix / 1,000 | 7 | 20 |
| Speed (median) | 5.9 sec | 2.3 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Personal data detection · French (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.1 pp. GPT-6 Luna is 3× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.71% | 99.62% |
| Error-free documents | 98.00% | 97.33% |
| Per 1,000 documents | $0.37 | $0.12 |
| To fix / 1,000 | 20 | 27 |
| Speed (median) | 5.3 sec | 2.4 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Contract clauses · English (US) (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.2 pp. GPT-6 Luna is 1.6× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 99.79% |
| Error-free documents | 100.00% | 98.67% |
| Per 1,000 documents | $0.32 | $0.20 |
| To fix / 1,000 | 0 | 13 |
| Speed (median) | 3.9 sec | 2.7 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Contract clauses · Czech (150 documents)
Both are equally accurate. GPT-6 Luna is 1.6× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 100.00% |
| Error-free documents | 100.00% | 100.00% |
| Per 1,000 documents | $0.42 | $0.27 |
| To fix / 1,000 | 0 | 0 |
| Speed (median) | 4.3 sec | 2.6 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Contract clauses · German (150 documents)
GPT-6 Luna is more accurate by 2.9 pp. GPT-6 Luna is 1.9× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 97.08% | 100.00% |
| Error-free documents | 76.00% | 100.00% |
| Per 1,000 documents | $0.52 | $0.27 |
| To fix / 1,000 | 240 | 0 |
| Speed (median) | 4.7 sec | 3.0 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Contract clauses · Spanish (150 documents)
Both are equally accurate. GPT-6 Luna is 1.5× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 100.00% |
| Error-free documents | 100.00% | 100.00% |
| Per 1,000 documents | $0.40 | $0.26 |
| To fix / 1,000 | 0 | 0 |
| Speed (median) | 4.1 sec | 3.1 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Contract clauses · French (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.5 pp. GPT-6 Luna is 1.6× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.85% | 99.33% |
| Error-free documents | 98.00% | 91.33% |
| Per 1,000 documents | $0.45 | $0.27 |
| To fix / 1,000 | 20 | 87 |
| Speed (median) | 5.2 sec | 3.2 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Document summaries · English (US) (150 documents)
Both are equally accurate. GPT-6 Luna is 3.4× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 100.00% |
| Error-free documents | 100.00% | 100.00% |
| Per 1,000 documents | $0.54 | $0.16 |
| To fix / 1,000 | 0 | 0 |
| Speed (median) | 8.8 sec | 2.9 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Document summaries · Czech (150 documents)
Both are equally accurate. GPT-6 Luna is 3.7× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 100.00% |
| Error-free documents | 100.00% | 100.00% |
| Per 1,000 documents | $0.78 | $0.21 |
| To fix / 1,000 | 0 | 0 |
| Speed (median) | 13.4 sec | 3.4 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Document summaries · German (150 documents)
Both are equally accurate. GPT-6 Luna is 3.2× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 100.00% |
| Error-free documents | 100.00% | 100.00% |
| Per 1,000 documents | $0.62 | $0.19 |
| To fix / 1,000 | 0 | 0 |
| Speed (median) | 9.8 sec | 3.3 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Document summaries · Spanish (150 documents)
DeepSeek V4.1 Flash is more accurate by 3 pp. GPT-6 Luna is 4.6× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 97.00% |
| Error-free documents | 100.00% | 82.00% |
| Per 1,000 documents | $0.88 | $0.19 |
| To fix / 1,000 | 0 | 180 |
| Speed (median) | 10.6 sec | 3.0 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Document summaries · French (150 documents)
DeepSeek V4.1 Flash is more accurate by 1.2 pp. GPT-6 Luna is 4× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 98.78% |
| Error-free documents | 100.00% | 92.67% |
| Per 1,000 documents | $0.78 | $0.20 |
| To fix / 1,000 | 0 | 73 |
| Speed (median) | 10.5 sec | 3.2 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Questions about a document · English (US) (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.1 pp. GPT-6 Luna is 2× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 99.22% | 99.11% |
| Error-free documents | 98.67% | 96.00% |
| Per 1,000 documents | $0.31 | $0.15 |
| To fix / 1,000 | 13 | 40 |
| Speed (median) | 4.2 sec | 3.1 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Questions about a document · Czech (150 documents)
Both are equally accurate. GPT-6 Luna is 1.7× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 100.00% |
| Error-free documents | 100.00% | 100.00% |
| Per 1,000 documents | $0.38 | $0.23 |
| To fix / 1,000 | 0 | 0 |
| Speed (median) | 3.5 sec | 3.1 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Questions about a document · German (150 documents)
DeepSeek V4.1 Flash is more accurate by 2 pp. GPT-6 Luna is 1.6× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 98.00% |
| Error-free documents | 100.00% | 91.33% |
| Per 1,000 documents | $0.33 | $0.20 |
| To fix / 1,000 | 0 | 87 |
| Speed (median) | 3.0 sec | 3.2 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Questions about a document · Spanish (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.3 pp. GPT-6 Luna is 1.8× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 99.67% |
| Error-free documents | 100.00% | 98.67% |
| Per 1,000 documents | $0.35 | $0.19 |
| To fix / 1,000 | 0 | 13 |
| Speed (median) | 3.3 sec | 2.9 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.
Questions about a document · French (150 documents)
DeepSeek V4.1 Flash is more accurate by 0.4 pp. GPT-6 Luna is 1.8× cheaper.
| Metric | DeepSeek V4.1 Flash | GPT-6 Luna |
|---|---|---|
| Correct fields | 100.00% | 99.56% |
| Error-free documents | 100.00% | 98.00% |
| Per 1,000 documents | $0.36 | $0.20 |
| To fix / 1,000 | 0 | 20 |
| Speed (median) | 3.6 sec | 3.0 sec |
| Valid answers | 100.00% | 100.00% |
Free with login
Accuracy field by field and the documents where the models differ.
Just your email, no password.